Mouse YOLO Factory

Generate simulated scratch defects, run YOLO model inference with auto-labeling, and merge mouse product defect image datasets with version control.

Security Scan

The skill's code appears to match its stated purpose (generating synthetic scratches, running YOLO inference, and merging datasets) but there are incoherent or risky implementation choices (hard-coded local paths for logs, missing dependency declarations, and SKILL.md using absolute Windows paths) that you should review before installing or running it.

Purpose & Capability

The code files implement scratch generation, YOLO inference (auto-labeling), and dataset merging, which aligns with the skill description. However the SKILL.md uses absolute Windows paths (D:/aiagent/aiagent_for_Mouse_Python_code/...) and the code itself creates a hard-coded RAG DB directory at D:/aiagent/rag_database — these hard-coded paths are not declared in metadata and may not match where the skill will be run.

Instruction Scope

SKILL.md instructs running scripts via absolute paths on D:, implying the agent or user should store/execute code there; the runtime code writes detection logs (JSONL) to a hard-coded local RAG path and writes/creates dataset folders and labels. While this is expected for dataset tooling, the instructions give no warning about these file writes and assume a Windows D: layout — this is scope creep relative to a simple 'run model' description and could overwrite or create files in unexpected locations.

Install Mechanism

There is no install spec (instruction-only + code files bundled). However the Python code depends on heavy native libraries (ultralytics, torch, torchvision, cv2/opencv, numpy, pandas) which are not declared. Users may attempt to run the scripts without these dependencies; installing them can be non-trivial and may require compiling native code. Absence of dependency declarations is an operational risk but not necessarily malicious.

Credentials

The skill requests no environment variables or credentials (which is good), but it writes logs to and creates directories under a hard-coded path (D:/aiagent/rag_database) and uses file-system locations for datasets. The skill does not declare these required config paths in metadata. There's no network exfiltration code, but local log files may include detection summaries — review these if they may contain sensitive image identifiers.

Persistence & Privilege

The skill does not request 'always: true' and does not modify other skills or global agent configuration. Its persistence is limited to creating directories and writing files (datasets, labels, and a local RAG JSONL log) within the file system; this is expected for dataset tooling.

What to consider before installing

What to check before installing or running: - The code implements the stated features, but the SKILL.md commands and some internal paths are hard-coded to D:/... — make sure you understand and relocate those paths to directories you control before running. - The scripts expect heavy Python dependencies (torch, ultralytics, torchvision, opencv, pandas, numpy). The skill metadata does not declare these — install them in a controlled environment (preferably a virtualenv or container). - The inference script writes detection logs to D:/aiagent/rag_database/detection_logs.jsonl and writes output images/labels into dataset folders. Inspect the logs if they may contain sensitive filenames or metadata. - Run the code first in an isolated environment (container or VM) and review/modify the hard-coded paths and any file-write locations. Search the code for any other absolute paths before trusting it with production data. - If you need tighter guarantees, ask the publisher to: remove hard-coded paths, declare required dependencies, and document exactly what files will be created/modified and where. - Confidence is medium: nothing in the code indicates network exfiltration or obfuscated/malicious behavior, but the path assumptions and missing dependency declarations are implementation issues that could lead to accidental data exposure or file overwrites.

這是一個專為老鼠 (Mouse) 產品瑕疵檢測開發的 YOLO 整合技能。支援瑕疵影像生成、自動化模型推論標記，以及資料集的合併與版本管理。

Metadata

• id: mouse-yolo-factory
• kind: package
• label: Mouse YOLO Factory
• owner: Alex Ho

Usage

1. 瑕疵生成 (Scratch Generation)

在原始影像上模擬生成劃傷 (Scratch) 瑕疵。 python D:/aiagent/aiagent_for_Mouse_Python_code/Mouse_produce_scratch.py --input <input_dir> --output <output_dir>

2. 模型推論與自動標記 (Auto-Labeling)

使用現有模型進行推論，並將結果儲存為 JSON/YOLO 格式。 python D:/aiagent/aiagent_for_Mouse_Python_code/drawbox_and_dataset_savejson_with_model.py --model <model_path> --img_size <size> --conf <threshold> --source <img_path>

3. 資料集合併與融合 (Dataset Merge)

將新標記好的資料併入全域資料庫。 python D:/aiagent/aiagent_for_Mouse_Python_code/datatool.py --new_data <new_path> --yolo_db <db_root> --desc <description>

Use when

• 使用者想要透過演算法在老鼠圖片上「產生」、「模擬」或「製作」劃傷瑕疵時。
• 使用者需要使用現有的 YOLO 模型對新圖片進行「推論」、「預標記」或「自動框選」時。
• 使用者提到「合併資料」、「融合數據集」、「建立新版本」或「將新資料歸檔」時。

Don't use when

• 進行硬體自動化測試（如 ADB 或 PLC 控制）時。
• 僅需進行單純的文件移動或重新命名，不涉及資料集邏輯時。